How real is the random censorship model in medical studies?

TL;DR

This paper examines the validity of the random censorship model in medical survival studies, emphasizing the need for evidence of its presence to avoid biased survival estimates.

Contribution

It differentiates types of censoring in medical studies and highlights the importance of verifying the random censorship assumption with empirical evidence.

Findings

Random censorship assumption is often presumed but not always validated.

Blindly assuming the model can bias Kaplan-Meier estimates.

Differentiation of dropout and administrative censoring is crucial.

Abstract

In survival analysis the random censorship model refers to censoring and survival times being independent of each other. It is one of the fundamental assumptions in the theory of survival analysis. We explain the reason for it being so ubiquitous, and we investigate its presence in medical studies. We differentiate two types of censoring in medical studies (dropout and administrative), and we explain their importance in examining the existence of the random censorship model. We show that in order to presume the random censorship model it is not enough to have a design study which conforms to it, but that one needs to provide evidence for its presence in the results. Blindly presuming the random censorship model might lead to the Kaplan-Meier estimator producing biased results, which might have serious consequences when estimating survival in medical studies.